x/tools/gopls: support completing to Elem() type for slices, etc.

[muirdm]

In the example:

var f func() string
var s []string
var a [1]string
var ch chan string
var m map[int]string

var _ string = <>

It would be nice if <> preferred s, a, ch, and m. We already prefer f. The completion candidate should include the appropriate expression to access the element type, such as s[<>] for indexable things, or <-ch for channels.

[findleyr]

Sounds good to me. We could theoretically go further (e.g. func() []string), but that's bound to get confusing. One degree of indirection feels about right.

[muirdm]

We could theoretically go further (e.g. func() []string),

I actually have a diff to do that as well. It searches up to three levels of indirection. It required some medium changes to record a proper ordering of the indirections (currently they are represented as separate bools). I'm still playing with it - will keep you apprised.

[gopherbot]

Change https://golang.org/cl/323449 mentions this issue: lsp/completion: reorganize how we track candidate type mods

[gopherbot]

Change https://golang.org/cl/323451 mentions this issue: lsp/completion: support completing to Elem() types

[gopherbot]

Change https://golang.org/cl/323450 mentions this issue: lsp/completion: search deeper for candidate type mods